File size: 5,573 Bytes
e4ba804 d958e80 17ea45c b7bd7bd 17ea45c ed03fc1 17ea45c 4a2e5e1 17ea45c d958e80 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 |
---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-4B-Thinking-2507
library_name: transformers
---
## Latest News
* [2026-01-12]🚀🚀🚀 We have open-sourced **AgentCPM-Explore**, an agent foundation model with only **4B parameters**, together with its **entire training and inference infrastructure**. AgentCPM-Explore has successfully entered **8 classic long-horizon agent benchmarks**, including **GAIA,HLE, and BrowserComp**. AgentCPM-Explore achieves **SOTA performance at the same parameter scale** and demonstrates its **accurate deep research capabilities**, effectively breaking the performance bottleneck for **on-device agents**.
## Overview
Key highlights of AgentCPM-Explore include:
- The **first full-parameter 4B agent model** to rank on **8 long-horizon and complex agent benchmarks**, including **GAIA, HLE, and BrowserComp**, in the on-device setting.
- Capable of **over 100 rounds of continuous environment interaction**, supporting **multi-source information cross-validation**, **dynamic search strategy adjustment**, and **real-time verification of up-to-date information**, enabling sustained deep exploration until task completion.
- **Fully open-sourced end-to-end**, including (1) **AgentRL**, a fully asynchronous reinforcement learning framework for agent training, (2) **AgentDock**, a unified management and scheduling platform for tool sandboxes, (3) **AgentToLeaP**, a one-click evaluation platform for agent tool-learning capabilities. These components collectively support **community collaboration and custom extensibility**.
We elaborate on the entire construction pipeline of AgentCPM-Explore on [GitHub](https://github.com/OpenBMB/AgentCPM).
## Experimental Results
<table>
<thead>
<tr>
<th>Model</th>
<th>GAIA (text-only)</th>
<th>BrowseComp</th>
<th>BrowseComp (ZH)</th>
<th>HLE</th>
<th>Frames</th>
<th>WebWalker</th>
<th>Seal-0</th>
<th>Xbench-DeepSearch</th>
</tr>
</thead>
<tbody>
<tr>
<td colspan="9"><strong>Closed-Source Models</strong></td>
</tr>
<tr>
<td>Claude-4.5-sonnet</td>
<td>71.2%</td>
<td>19.6%</td>
<td>40.8%</td>
<td>24.5%</td>
<td>85.0%</td>
<td>/</td>
<td>53.4%</td>
<td>66.0%</td>
</tr>
<tr>
<td>Gemini Deep Research</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>26.9%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>/</td>
</tr>
<tr>
<td>DeepSeek-V3.2</td>
<td>63.5%</td>
<td>67.6%</td>
<td>65.0%</td>
<td>40.8%</td>
<td>80.2%</td>
<td>/</td>
<td>38.5%</td>
<td>71.0%</td>
</tr>
<tr>
<td>MiniMax-M2</td>
<td>75.7%</td>
<td>44.0%</td>
<td>48.5%</td>
<td>31.8%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>72.0%</td>
</tr>
<tr>
<td>OpenAI-GPT-5-high</td>
<td>76.4%</td>
<td>54.9%</td>
<td>65.0%</td>
<td>35.2%</td>
<td>/</td>
<td>/</td>
<td>51.4%</td>
<td>77.8%</td>
</tr>
<tr>
<td>GLM-4.6</td>
<td>71.9%</td>
<td>45.1%</td>
<td>49.5%</td>
<td>30.4%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>70.0%</td>
</tr>
<tr>
<td>Kimi-Researcher</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>26.9%</td>
<td>78.8%</td>
<td>/</td>
<td>36.0%</td>
<td>69.0%</td>
</tr>
<tr>
<td>Seed-1.8</td>
<td>87.4%</td>
<td>67.6%</td>
<td>81.3%</td>
<td>40.9%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>/</td>
</tr>
<tr>
<td colspan="9"><strong>Open-Source Models</strong></td>
</tr>
<tr>
<td>MiroThinker 8B</td>
<td>66.4%</td>
<td>31.1%</td>
<td>40.2%</td>
<td>21.5%</td>
<td>80.6%</td>
<td>60.6%</td>
<td>40.4%</td>
<td>60.6%</td>
</tr>
<tr>
<td>Tongyi DeepResearch 30B</td>
<td>70.9%</td>
<td>43.4%</td>
<td>46.7%</td>
<td>32.9%</td>
<td>90.6%</td>
<td>72.2%</td>
<td>/</td>
<td>75.0%</td>
</tr>
<tr>
<td>ASearcher QWQ 32B v2</td>
<td>58.7%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>74.5%</td>
<td>/</td>
<td>/</td>
<td>51.1%</td>
</tr>
<tr>
<td>iterresearch-30B-A3B</td>
<td>72.8%</td>
<td>37.3%</td>
<td>45.2%</td>
<td>28.8%</td>
<td>71.0%</td>
<td>/</td>
<td>39.6%</td>
<td>/</td>
</tr>
<tr>
<td>WebSailor-V2-30B-A3B (RL)</td>
<td>74.1%</td>
<td>35.3%</td>
<td>44.1%</td>
<td>30.6%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>73.7%</td>
</tr>
<tr>
<td>WebLeaper-30B-A3B-RUC</td>
<td>73.2%</td>
<td>38.8%</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>/</td>
<td>48.6%</td>
<td>72.0%</td>
</tr>
<tr>
<td>WebDancer (QWQ-32B)</td>
<td>51.5%</td>
<td>3.8%</td>
<td>18.0%</td>
<td>/</td>
<td>/</td>
<td>47.9%</td>
<td>/</td>
<td>38.3%</td>
</tr>
<tr>
<td>⭐ <strong>AgentCPM-Explore 4B</strong></td>
<td>63.9%</td>
<td>25.0%</td>
<td>29.0%</td>
<td>19.1%</td>
<td>82.7%</td>
<td>68.1%</td>
<td>40.0%</td>
<td>70.0%</td>
</tr>
</tbody>
</table> |